Using Markov chain analysis to study dynamic behaviour in large-scale grid systems

摘要: In large-scale grid systems with decentralized control, the interactions of many service providers and consumers will likely lead to emergent global system behaviours that result in unpredictable, often detrimental, outcomes. This possibility argues for developing analytical tools allow understanding, prediction, complex behaviour order ensure availability reliability computing services. paper presents an approach using piece-wise homogeneous Discrete Time Markov chains provide rapid, potentially scalable, simulation systems. approach, previously used other domains, is here model dynamics this a chain first represented reduced, compact form. can then be perturbed produce alternative execution paths identify scenarios which performance degrade or anomalous occur. The expeditious generation these allows prediction how larger react failures high stress conditions. Though computational effort increases proportion number modelled, cost shown far less than detailed testbeds. Moreover, unaffected by size being expressed terms workload resources, adaptable are non-homogenous respect time. provides examples application approach.